Making the Robust Tensor Estimation Approach: "RESTORE" more Robust

Introduction: The Robust Estimation of Tensors by Outlier Rejection (RESTORE) [1] has been demonstrated to be an effective method for improving tensor estimation on a voxel by voxel basis in the presence of artifactual data points in the diffusion weighted images (DWIs). However, the RESTORE method that combines robust regression and outlier rejection techniques relies on data redundancy [2]. The estimated parameters may not be reliable if the data set does not have enough good data points to correctly identify outliers, or when too many data points have been excluded from the fitting. Moreover, the criteria used for outlier identification in DWIs may not be reliable when applying the same rule to non-DWIs due to intrinsic higher physiologic noise in T2 weighted images. Instabilities on tensor derived quantities have been observed in noisy clinical brain data which causes difficulties in statistic DTI analysis. In this paper, two practical constraints are introduced to further improve the robustness of the RESTORE algorithm.